MOSAIC Books · AI Capture

A receipt workbench built for company firms

For teams serving SME clients with high receipt volume. MOSAIC Books keeps OCR, review and posting in one controlled workflow—auditable totals, staff in the loop.

Sound familiar?

  • Clients dump receipts every month—junior staff live in data entry
  • Mixed formats: retail, taxi, restaurants—raw OCR still means hunting for totals
  • Partners and clients ask: can we stay off public cloud?
  • You want review before Xero—not “scan and auto-post”

How MOSAIC Books helps your firm

Upload into the workbench

Scanner, camera or file upload (online demo: upload only). Production fits daily firm workflows.

Fields that match bookkeeping

Merchant, date/time, total, payment—totals from OCR settlement lines, aligned with how staff review.

One review screen

Edit then confirm. Reporting, reconciliation and confirm stay with your staff and system—not a cloud AI.

Post & Xero after confirm

Local books update after confirm; Xero push only for items you select—never auto-post.

A typical day at the firm

  1. Morning: client receipts arrive (email, WhatsApp, scans)
  2. Batch upload → OCR → review screen
  3. Supervisor spot-check or rule-based review
  4. Confirm posting; push to Xero when needed
  5. Month-end: reports and exports from confirmed data

Two deployment tiers

Standard · OCR + rules

Best for: high volume, cleaner layouts, controlled hardware cost

Windows host + Mini G (RapidOCR CPU/GPU). Without on-prem AI we stay on the standard tier—no fake “AI” badge.

Smart · OCR + on-prem AI

Best for: messy layouts—merchant/time/payment from OCR text

Local AI models on your machine; totals still OCR-only. Client receipt data is never uploaded online—compute stays in your environment.

Who it's for

  • Company firms serving multiple SME clients
  • Bookkeeping BPO and receipt hubs
  • Teams piloting digitisation + review before Xero
  • Clients who need on-prem options—not all receipts to public OCR clouds

v1: no client chat; no expense policy engine via RAG (limits use config/SQL, not knowledge base).

Control & compliance your partners expect

  • No posting before review confirm
  • Totals are an accounting field—OCR settlement only, no model arithmetic
  • Mini G License is machine-bound; separate from demo auth codes
  • No client chat in v1—fewer misuse and compliance surprises
  • Online demo is for trial; production data should be on-prem / dedicated

About the online trial

Upload-only demo: enter an auth code, up to 10 attempts (failed runs count too). Evaluate OCR and field extraction—not every production feature.

Open online demo

FAQ for company firms

How accurate is it?

Tuned for common HK-style receipts; totals from OCR settlement lines. Edit on review—confirmed data is source of truth.

Auto-posting?

No. Review confirm first; Xero only after you select confirmed items.

GPU required?

No for standard tier (OCR + rules). Smart tier runs local AI models on your machine.

Cloud only?

Production can be a Windows appliance + local Mini G. Smart-tier AI stays on-prem—client receipt data is never uploaded online. Online demo is for trial only.

vs generic OCR?

We’re a workbench: extract → review → confirm → post/Xero—not just text output.

Pricing?

Quoted by firm size, tier and deployment. Book a demo to discuss.

Try 10 receipts online first—chat if you’re interested

Use an auth code to try. Like what you see? WhatsApp or email us to talk deployment.